On 12.09.18 - 09:50, Florian Pritz via arch-devops wrote:
I don't know if we really want or need mirrorbrain. It's a great fit for projects with few big files which are downloaded by many people upon a new release. If we were to use mirrorbrain, the pacman mirror list would only contain the redirector URL, correct? That would then mean that pacman asks the redirector for each individual package which I think generates way too much unnecessary load compared to our current system. Also, right now people can update without having to connect to a central server and I don't like the idea of adding a SPOF for something that already works rather well.
Mirrorbrain sounds fine for our ISOs, but I don't know if those get enough downloads (via HTTP) to warrant setting it up.
Florian
I suppose you could have multiple mirrorbrain hosts with synced config/database (pgsql streaming replication works quite well) and then there could be multiple entries in the client's mirrorlist again. Otherwise it could still be the same FQDN but as a DNS RR with multiple records and a somewhat low TTL of 300s-ish? The project sounds quite nice and well suited for the job and should be able to centralize/unify or current situation, but I think we have some other things with a higher priority than changing/optimizing something that is working reasonably well currently. Cheers, Thore -- Thore Bödecker GPG ID: 0xD622431AF8DB80F3 GPG FP: 0F96 559D 3556 24FC 2226 A864 D622 431A F8DB 80F3